基于眼动轨迹分析的BPPV诊断算法研究

Research on BPPV Diagnosis Algorithm Based on Eye Movement Trajectory Analysis

近年来,以机器学习模型辅助临床诊断已成为智慧医疗领域的一大研究热点。在良性阵发性位置性眩晕(Benign Paroxysmal Positional Vertigo,BPPV)的临床诊断上,眼动视频的解释往往是由医生观测得出的,但直接观测诊断的局限性在于难以捕捉细微的眼动特性而容易导致误诊。对此,论文提出了一种基于时序轨迹的BPPV诊断模型,将基于深度学习的目标检测器和基于时序数据分析的分类器相结合以实现BPPV诊断。具体地,该模型对眼动视频进行眼球目标检测,以提取眼球运动的时序轨迹,并结合数据增强对训练样本进行扩充,以准确分类并得到更好的诊断结果。实验结果表明,论文提出的模型可有效提取眼动时序轨迹,并在BPPV诊断上取得良好性能。

In recent years,it has become a hot research topic to utilize machine learning models to assist clinical diagnosis in the field of intelligent medicine.In the clinical diagnosis of benign paroxysmal positional vertigo(BPPV),the interpretation of eye movement video is mostly carried out by doctors.However,the limitation of direct eye observation which is difficult to capture subtle eye movement characteristics and may lead to misdiagnosis.This paper proposes a diagnosis model of BPPV based on time sequence trajectory,which combines the deep learning based object detector and the time series analysis based classifier for the diagnosis of BPPV.Specifically,the proposed model detects the eye object and extract the time series trajectory of the eye movement.With the data augmentation,the training samples are expanded,so as to obtain better classification and diagnostic results.Experimental re⁃sults show that the proposed model can effectively extract the time series trajectories of eye movement,and achieve high-quality per⁃formance in the diagnosis of BPPV.